Over/under probability at the book line for every player on the slate: an XGBoost regression projects the mean from his last eight games and the matchup, his recent spread sets the width, and a z-score scores the line. Graded 462-451 (50.6%) at the play tier 2025-26 season.
Graded record, Mar 12, 2026 – Jun 13, 2026. Full NBA performance page — every market, every pick counted.
Projections are published for every game day of the season and graded against the box score the next morning. Until opening night, here is what the model hit last season (Mar 12, 2026 – Jun 13, 2026, 79 graded days, playoffs included) — the same counters as the performance page.
| Market | Play rule | Record | Hit rate |
|---|---|---|---|
| Moneyline | model ≥ 60% | 156-41 | 79.2% |
| Spread | projected margin ≥ 5 | 155-52 | 74.9% |
| Totals | strong play, 60–70% | 58-46 | 55.8% |
| Points | model ≥ 65% | 462-451 | 50.6% |
| Rebounds † | model ≥ 65% vs its own line | 203-90 | 69.3% |
| Assists | model ≥ 65% | 450-337 | 57.2% |
| Double-double | model ≥ 60% | 93-53 | 63.7% |
| Triple-double | model ≥ 30% | 11-10 | 52.4% |
Green clears the -110 break-even of 52.4%; amber does not. Hit rates are historical and not a guarantee. † Rebounds is graded against a line the model sets from the player's own recent average (no book carries one), so its hit rate is not comparable to a market price. How the props are graded, and how to find value.
Last week's graded record — wins and losses — the plays for the week ahead, and a subscriber discount code. One email, no daily noise, unsubscribe in a click.
Points is the noisiest of the three big NBA prop markets. A player's scoring swings with shot volume, foul trouble, blowouts and rotation changes, so a single projected number tells you little. The model builds a distribution instead: an XGBoost regression projects the mean from his last eight games (minutes and points per minute), rest, back-to-back, home/away and the opponent's defensive rating and pace; the spread of those eight games, floored and widened by a calibration factor, sets the width; and a z-score against the line gives the probability that he finishes over it.
A play is badged only when one side reaches 65%. That threshold is the same one the grader uses, so the hit rate on this page is the hit rate of exactly what was badged — not of a curated subset.
Last season the points model graded roughly even at the 65% tier — around 50%, which does not beat the -110 break-even of 52.4%. Rebounds and assists did; points did not. That number is on the page because a prop board that hides its own record is not worth reading. The practical use of the points board is the distribution: the alt-line probabilities and the last-10 chart show where a line sits inside a player's recent range, which is more useful than the direction call at the main line.
Where points DOES carry signal is in-game: the live points tracker re-projects every 60 seconds from the box score and the checkpoint models at Q1, Q2 and Q3 grade well above the pre-game number. If you want the points market, that is the tool.
Each card shows the book line, the model's over/under split, the last-10 average against the line, and — under Details — the alt lines with their over probabilities and a bar for each of the last ten games (green cleared the line, red did not). A player whose last-10 bars are all green at a line two points below his average is a different bet from one whose bars alternate, even if both print 65%.
Points, rebounds, assists, double-doubles and triple-doubles, plus live in-game trackers. MLB, NFL and NHL included.
Start your free 3-day trial